Can America’s Tech Giants Get Together & Start a Chipmaker?
I was listening to a podcast, and the guys on the show started talking about TSMC. They noted that the majority of TSMC's customers are American tech giants - Apple, Nvidia, AMD, so on ... So they said, what if those American companies came together to invest in and buy chips from a startup American foundry? They liked the idea. But discounting the anti-trust kerfuffle that will inevitably follow, I must say that I have heard something like it before in history.
In the second half of 1989, near the end of a decade of US-Japan semiconductor tensions, there was a major push for something similar. It failed. In this video, we look at US Memories, the American memory production consortium. ## Beginnings To understand what happened with US Memories, we need to learn the great memory shock of the 1980s - one of the most challenging times in American semiconductors. The challenge was in the arena of the industry's largest market at the time - Dynamic RAM memory, used as secondary memory in computers. Such computers are insatiable in their demand for more memory. Dynamic RAM's
high density and cheap commodity prices make it the preferred choice for system integrators putting together their products. We classify them in generations, with each subsequent generation being 4 times larger than the previous one. So 1 kilobit Dynamic RAM to 4 to 16, so on. ## The Japanese Challenge An American invented the concept, the late great Robert H. Dennard,
who recently passed away at the age of 91. And American companies turned the Dynamic RAM concept into an growing, profitable industry. First, Intel with the 1-kilobit Intel 1103 chip, released in 1970. And then came Mostek, which became a $350 million company by dominating the 4 and 16 kilobit generations. At the latter generation, Mostek had 75% share of the world market at its peak.
Mostek also pioneered the 16-pin industry standard for Dynamic RAM modules. Japanese semiconductor companies seized on this open format to enter the market. They leveraged a combination of superior manufacturing techniques and cheap money to grab share. By 1979 and the 64 kilobit generation, Hitachi, NEC and Fujitsu had caught up with the Americans. ## Market Crash Nevertheless, in 1983, Americans still owned 60% of the market. Seeing nothing but goodness ahead in the memory market thanks in part to the exploding PC industry, the Americans invested heavily in new capacity only to be punched in the face by a massive downturn.
The economy slowed down in late 1984 and 1985, which then caused the bottom to fall out of the Dynamic RAM market. During the worst of it, 256-kilobit chips were selling for $2 a unit. 64-kilobit chips less than a dollar. The whole semiconductor industry lost $5 billion. $4 billion of which was lost by the Japanese. But those Japanese players survived, thanks in part to the financial wherewithal of their electronics arms and another part to what was termed as "long term thinking".
The Americans were not so lucky. Before the crunch, there were 11 US semiconductor companies making Dynamic RAM memory: AMD, AMI, Fairchild, Intel, ITT, Micron Technology, Mostek, Motorola, National Semiconductor, Texas Instruments and Zilog. By 1986, just three of those remained: Micron, TI and Motorola. And TI and Motorola were qualified cases. Texas Instruments' memory facility stayed open really just
for technology R&D. Motorola exited in 1984, but has kept an eye on the market ever since. So that really just left Idaho-based Micron as the last true American memory-maker. I covered Micron in a previous video. In other words, the Japanese now reigned supreme. ## What Is To Be Done? As this played out, everyone in the US semiconductor industry was asking what to do defend itself? Everyone was divided on this. The Semiconductor Industry Association
or SIA was deeply divided. As early as 1980, Motorola wanted to file a dumping complaint against the Japanese, but the board twice voted against it. The problem was that a sizable portion of the SIA was made up of chipmakers with computer-making arms like IBM. IBM was then enjoying its PC Boom. And though it wanted a healthy market, it opposed things like quotas or tariffs that might limit a critical input for its lucrative PCs. The Reagan Administration was also divided. The cabinet was committed to the idea of free
trade. Moreover, the implosion of the American DRAM market happened so quickly - basically 2-3 years - that several argued that it would be like shutting the barn door after the horse bolted. In the end, Micron launched their own lawsuit against the Japanese, and the Reagan administration backed it. Though I should note that it was Micron's own late 1984 price cut that kicked off the worst of the 1985 slump. Micron's dumping charges led to others, like an EPROM dumping petition by Intel, AMD and National Semiconductor. These were motivated by the discovery of a Hitachi "10% memo", where the company urged its distributors to undercut competitors' prices by 10% no matter the economics.
Concerns about a burgeoning trade war led to the 1986 Semiconductor Trade Agreement. Japan agreed to self-limit what it sold into America as well as to somewhat open its market to American-made memory chips. ## A Chip Shortage! What happened next might not surprise people. The 1986 agreement did nothing to bring back the dead. The horse had indeed bolted from
the barn. The Japanese - facing down tariffs in 1987 - cut their production of 256K chips while they upgraded their fabs to prep for 1 megabit. Then starting in mid-1988, the economy surged. Demand came back, catching the Japanese by surprise. 256K supply was down and 1 megabit had not yet ramped to full scale. Thus, DRAM prices in the third quarter of 1988 surged to unprecedented heights.
People were now talking about a chip shortage, a crisis in 1988. Jack Tramiel, then the CEO of Atari, said in June 1988: > Nine months ago, DRAMs sold for $1.50. Today on the spot market, those same chips cost $6. It didn't take the Japanese long to figure out that they can get a higher price for selling fewer chips. They don't mind profiting from American stupidity. Micron turned record profits during the shortage. But it wasn’t so fun for the larger PC, computer, and consumer electronics industries. DRAM was their oxygen and despite
the US fabs’ attempts to scale up, there was not enough to go around. Companies from Atari to Apple to Sun could not ship product or had to pay more because of memory shortages. The average PC or Macintosh needed 512K at a minimum. Apple's Macintosh SE got pushed back a year. Sun's workstations like the 386i or 3/60 suffered backlogs six months long. It was not just computers either. Laser printers, scanners, fax machines, barcode readers and more. All of these used Dynamic RAM modules for their products. One thing that everyone had plenty of, however, was blame. Blame went to the Japanese for "stealing" and then "strangling" the
market by first dumping the market and then extorting the Americans after winning it. Blame went to the 1986 Trade Agreement - and the American semiconductor makers that brought it on - for limiting American access to a critical computer commodity. Blame also went to the American semiconductor companies for exiting the Dynamic RAM market, and not fighting to stay the course through the market turbulence.
Motorola and National Semiconductor did think about re-entering the market, but questioned the wisdom of investing so many millions of dollars so long the Japanese can come back again. So, continued protections, which the government declined to retain for the long term. ## DRAM OPEC The shortage began to subside in a few months, starting with the second half of 1989. Spot prices remained somewhat elevated, but were trending down. The Japanese and Koreans ramped up their memory supply. And importantly for later in our story,
the American computer-makers struck custom deals with the Japanese overseas. But the Great Chip Shortage of 1988 got people talking about America re-entering the market. The dependency on Japanese memory chips then terrified the computer-makers and the remaining semiconductor makers too. The reason being that, first, DRAM manufacturing excellence has trickle-down effects because it is so repetitive and dense. Second there were serious concerns that the Japanese would leverage their memory dominance to make inroads into other sectors like ASICs. As in they would say,
"If you want to buy our DRAMs, you need to buy our other products too". Bundling. People in the press and industry talked about how the American electronics industry should not be subject to what was called a "DRAM OPEC". There were rumors of Apple Computer and Sun Microsystems possibly starting their own DRAM fabs in Silicon Valley. Tramiel in June 1988 also publicly announced that Atari hoped to buy or make their own semiconductor fab too. Several people in the industry saw this as an opportunity. One of the lessons that the Americans learned from Japan's successes in the late 1970s was the use of collaboration structures within an entire industry.
Such lessons led directly to Sematech, a research consortium which received government and industrial money to research certain technologies. As Sematech came together in 1988, the SIA started thinking about maybe a similar consortium for manufacturing. The idea was that if any no single US chip company was willing to stick its neck out and get back into the DRAM market, then maybe a collective effort could. In March 1989, the SIA and the American Electronics Association, which represents the computer makers, began working on an effort. ## US Memories In June 1989, the SIA announced the formation of a cross-industry joint production venture between America's largest computer and chip makers. The company was called US Memories.
The idea was backed by seven American tech giants. They were IBM, AMD, Digital Equipment Corporation or DEC, Hewlett-Packard, Intel, LSI Logic, and National Semiconductor. Three of these - IBM, DEC, and HP - were major computer manufacturers. Wilf Corrigan of LSI Logic served as chairman. The CEO was Sanford Kane, a long-time senior executive in IBM's semiconductor division. He joined IBM right out of college in 1962 and had worked there for 27 years.
After participating in the founding of Sematech, Kane was asked to lead the US Memories effort as its only full time employee - though he had up to 25 workers under him "borrowed" from the original seven partners. The goal was to have a 4 megabit Dynamic RAM fab up and running by the end of 1991. About a year and a half away. The product and process technology would be licensed from IBM,
which was then already making DRAMs but only for its internal usage. Kane saw the company as a cooperative. Whoever joined and invested can buy a corresponding percentage of their DRAM needs at a good price. The US Memories concept was initially costed at about a billion dollars, or about $2.5 billion today. The seven companies contributed about $100,000 each for admin expenses. Kane's job was to find investors for the rest of the money. If he managed to do it,
then US Memories was projected to be a billion dollar company by 1993. ## Anti-trust The US Memories consortium's backers - which included three of America's largest computer manufacturers - evoked anti-trust concerns. Sematech was an R&D cooperative of all of America's biggest semiconductor makers. It and others like it were only possible thanks to the National Cooperative Research Act of 1984 or NCRA, which reduced the potential antitrust liabilities of such an arrangement. US Memories was different in that it actually manufactured products for profit. It was the most public such setup since the Toyota-GM NUMMI joint venture some
six years earlier in 1983. That joint venture passed antitrust scrutiny but with conditions. In Congressional hearings in 1989, Sandy Kane testified that US Memories presented no anti-competitive concerns because by 1992 at its start, it still would only have 10% of a multi-billion dollar market. He did not ask for funding from the government, but did seek some sort of legislative protection from anti-trust threats on joint production ventures like what the NCRA granted to R&D cross-industry ventures like Sematech. I asked Kane why he didn't try to get government money for US Memories - considering how military backing helped get Sematech off the ground. Kane said that a few Congressmen suggested it, but taking government money would have opened them up to far more serious antitrust scrutiny.
And unlike Sematech, US Memories was to be a profit-making commercial venture. ## Opposition Kane told me that he never personally heard serious opposition to the idea while campaigning for US Memories. But a few people did question whether it so mattered that 5, 8, or even 10% share of the Dynamic RAM market was being produced on American soil. Scott McNealy, cofounder and chairman of Sun Microsystems, later said it was like "pouring water on a desert in the hopes of regenerating a forest."
He also said, "We have contracts with worldwide suppliers ... as long as I have a reasonably consistent supply at a price, I don’t care where it comes from." You can get a hint of why Kane thought that the computer-makers lacked "strategic thinking". Considering that all of them are gone except Apple, maybe he wasn't wrong. Anyway, the most public voice speaking out against US Memories was TJ Rodgers of Cypress Semiconductor, which made SRAMs. Rodgers had also spoken out against Sematech, calling it a limiter on smaller, more entrepreneurial companies. Rodgers is a bit of a character, always willing to give a spicy quote. In an interview,
he claimed Cypress can also do what US Memories did if it got the same technology transfer and volume purchase commitments from IBM. ## Weak Demand In the end, however, antitrust and industry opposition were less the problem than a lack of commitment from the potential customers. Kane had hoped that a few of the major computer makers - like Apple, Sun, or Unisys - would join. Unfortunately, they turned it down. Apple - then in its clown phase - ended up buying a bunch of DRAM at the worst possible time.
Compaq, then a major computer-maker, also could not commit. Even HP, one of the original seven, hedged too. So many just went on and on without a formal commitment. There were several concerns. First, the worst of the memory chip shortage
had passed and DRAM prices had started to fall. Companies were no longer as excited to put up $5 to 50 million of their own money when cheaper chips were now on the market. Second, there were concerns about a startup company being able to rapidly scale up into a 2,000-person enterprise, while also setting up a full-scale high volume wafer fab. Can they really deliver chips in less than two years? I felt the timeline was ambitious but doable. After all, the technology was IBM's, and they were already using it.
And third, they were again concerned about the Japanese. Nobody said it outright, but they did allude to it. During the shortage, the US computer makers struck long term supply agreements with the Japanese. They were worried that if they invested in US Memories,
then the Japanese would cut them off - leaving them short until US Memories was up and running. In other words, a threat - albeit unspoken. ## The Big Meeting Time was running out. 4 megabits was reasonably leading edge, but the Japanese were already then pushing to 16 megabits. The end came at a pivotal meeting in Dallas with 11 electronics companies, including the original seven partners.
By now, Kane rejiggered the plan. The most critical thing was raising about $150 million of equity investment from the electronics industry, particularly the computer-makers. If that was possible, then Goldman Sachs - US Memories' financial advisor - reckoned that they can raise another $400-600 million in the form of equity and debt from Wall Street. Goldman said however, that those investors would require that the industry investors have to commit to buying more of US Memories' output.
So Kane put up a big poster board with what everyone in the room had committed for the venture - financially and in purchase commitments. Only IBM and DEC of the computer-makers had firmly committed, and that was not enough. So Kane asked if anyone wanted to add more. He later recalled, "The room got real quiet,
real fast ... I've been to livelier wakes". A few days later, January 15th, Kane - incredibly disappointed - announced that the US Memories concept cannot go forward. To many in the industry, the US Memories outcome showed a distinct lack of long-term strategic thinking in American companies. The NYTimes ran an Op-ed titled "US Memories: Too Short" Gordon Moore, Intel's co-founder, said: "It's not a very strong endorsement of getting together to solve problems" Another former electronics executive said, "This will surely come back to haunt us".
As predicted, the US never recovered the domestic market share in DRAM it once had during the glory days of Mostek. But the rise of the Korean memory giants and Intel's CPU monopoly largely settled the political fears of the US-Japan semiconductor trade tensions in the 1980s. Kane went on to bigger and greater things. Most notably to me, he led the spin off of the National Semiconductor fab in Israel to found Tower Semiconductor. He served as their founding CEO for about two years. ## Conclusion I want to thank Sandy Kane for taking the time to speak with me and answering my dumb questions. There are not many semiconductor veterans around and Sandy had many wonderful stories.
Can an arrangement like US Memories be competitive with TSMC, today? I asked him at the end of our conversation and he said no. Getting $750 million for US Memories was a lot, but then seemed achievable with the help of Wall Street. Imagine how many tens of billions must be raised in debt and equity to compete with TSMC.
SemiAnalysis estimated in April that Intel has potentially $116 billion on tap for its foundry efforts, and they are still not there yet. $116 billion split amongst the Magnificent 7 tech companies is $16.5 billion each. Even for them, that's a lot. The American computer-makers weren't even willing to do $50 million, or $120 million today, for US Memories.
And where is the technology coming from? US Memories licensed it from IBM at a good price. Logic ASIC chips customized for a customer's specific purpose - delivered affordably, on time, using the latest nodes, with good yield, and at immense, incomprehensible scale. The product and process technologies for that are not freely available. I can go on, but this video has gone long. But as the silicon world ramps up for another boom, the US Semiconductor Manufacturing Company idea lingers with me, but hanging right alongside it are the lessons of US Memories.
2024-09-25 21:25
